rename files when sending to s3 and match correctly when updating cocina

lib/dor/text_extraction/cocina_updater.rb

@@ -13,13 +13,14 @@ module TextExtraction
     # 2. If the file will overwrite an existing file in Cocina ensure that
     #    the file to be overwritten was generated by SDR and has not been
     #    corrected.
-    # 3. If the file already exists in the Cocina update it by first deleting
+    # 3. If the file already exists in the Cocina, update it by first deleting
     #    from the Cocina and then adding it again.
-    # 4. If the file does not exist in the Cocina already look find a resource
-    #    that contains the same filename "stem" and add the file there. e.g.
-    #    "abc123.xml" would be added to the same resource that contains
-    #    "abc123.tiff".
-    #
+    # 4. If the file does not exist in the Cocina already, look find a resource
+    #    that contains the same filename "stem" or the same filename with extension added to the end of the stem
+    #    and add the file there.
+    #    e.g. "abc123.xml" would be added to the same resource that contains "abc123.tiff"
+    #    and "abc123_tiff.xml" would be added to the same resource that contains "abc123.tiff"
+    #    see https://github.com/sul-dlss/common-accessioning/issues/1443 for the logic behind this
     # rubocop:disable Metrics/ClassLength
     class CocinaUpdater
       attr_reader :dro, :logger
@@ -149,14 +150,17 @@ def find_resource(path)
         end
       end
 
-      # Find a Cocina resource for the given path by looking for a resource that has a matching filename "stem".
+      # Find a Cocina resource for the given path by looking for a resource that has a matching filename "stem"
+      # OR a matching filename with an extension added to the end of the stem,
+      # e.g. "file1_tiff.xml" would be added to the same resource that contains "file1.tiff"
+      # see https://github.com/sul-dlss/common-accessioning/issues/1443
       # @param path {String} - the path to look for
       # @return {FileSet, nil} - the resource or nil if it is not found
       def find_resource_with_stem(path)
         file_stem = stem(path)
         structural.contains.detect do |resource|
           resource.structural.contains.detect do |file|
-            stem(file.filename) == file_stem
+            stem(file.filename) == file_stem || file.filename == extracted_filename_with_extension(path)
           end
         end
       end
@@ -240,8 +244,17 @@ def resource_type(path)
         end
       end
 
+      # get the stem of a file path removing the extension (e.g. '/user/path/file1.xml' => 'file1')
       def stem(path)
-        File.basename(path, '.*').to_s
+        File.basename(path, File.extname(path))
+      end
+
+      # get the full filename of a file path by extracting the extension from the last part of the filename (e.g. '/user/path/file1_mp4.vtt' => 'file1.mp4')
+      def extracted_filename_with_extension(path)
+        basename_with_extension_stem = stem(path)
+        extension_from_filename = basename_with_extension_stem.include?('_') ? basename_with_extension_stem.split('_', 2).last : ''
+        basename_without_extension_stem = basename_with_extension_stem.gsub("_#{extension_from_filename}", '')
+        "#{basename_without_extension_stem}.#{extension_from_filename}"
       end
 
       def file_identifier

lib/dor/text_extraction/speech_to_text.rb

@@ -57,8 +57,18 @@ def filenames_to_stt
       end
 
       # return the s3 location for a given filename
+      # NOTE: Due to https://github.com/sul-dlss/common-accessioning/issues/1443, we will rename this file when sending it to whisper
+      # to ensure whisper will always produce a unique output caption file for each input file provided (even if the basename is identical).
+      # This is because whisper currently only uses the basename to determine the output filename, which will cause a problem if we have two
+      # input files with the same basename but different extensions (e.g. "file.mp4" and "file.m4a").
       def s3_location(filename)
-        File.join(job_id, filename)
+        file_extension = File.extname(filename)
+        basename = File.basename(filename, file_extension)
+        # For example, "file1.mp4" becomes "file1_mp4.mp4" which will cause whipser to produce "file1_mp4.vtt" as the output,
+        # and "file1.m4a" becomes "file1_m4a.m4a" which will cause whisper to produce "file1_m4a.vtt" as the output, ensuring uniqueness for both.
+        # The cocina_updater will understand this and update the cocina structural metadata correctly matching the new filename to the original.
+        new_filename = "#{basename}_#{file_extension.delete('.')}#{file_extension}"
+        File.join(job_id, new_filename)
       end
 
       # return the job_id for the stt job, defined as the druid-version of the object

spec/lib/dor/text_extraction/speech_to_text_spec.rb

@@ -212,8 +212,8 @@
   describe '#s3_location' do
     let(:version) { 3 }
 
-    it 'returns the s3 filename key for a given filename' do
-      expect(stt.s3_location('text.xml')).to eq("#{bare_druid}-v#{version}/text.xml")
+    it 'returns the new s3 filename key for a given filename' do
+      expect(stt.s3_location('text.xml')).to eq("#{bare_druid}-v#{version}/text_xml.xml")
     end
   end
 

spec/robots/dor_repo/speech_to_text/update_cocina_spec.rb

@@ -62,6 +62,7 @@
 
     it 'runs the update cocina robot and sets the caption role' do
       new_cocina = test_perform(robot, druid)
+      expect(new_cocina.structural.contains.size).to eq 1
       new_json_file = new_cocina.structural.contains[0].structural.contains[1]
       new_vtt_file = new_cocina.structural.contains[0].structural.contains[2]
       expect(new_json_file.filename).to eq 'file1.json'
@@ -71,6 +72,82 @@
     end
   end
 
+  context 'with vtt files and input files with the same filename but different extensions' do
+    let(:structural) do
+      {
+        contains: [
+          {
+            type: Cocina::Models::FileSetType.file,
+            externalIdentifier: "#{bare_druid}_1",
+            label: 'Audio 1',
+            version: 1,
+            structural: {
+              contains: [
+                {
+                  type: Cocina::Models::ObjectType.file,
+                  externalIdentifier: "#{druid}_1",
+                  label: 'file1.m4a',
+                  filename: 'file1.m4a',
+                  size: 123,
+                  version: 1,
+                  hasMimeType: 'audio/mp4'
+                }
+              ]
+            }
+          },
+          {
+            type: Cocina::Models::FileSetType.file,
+            externalIdentifier: "#{bare_druid}_1",
+            label: 'Video 1',
+            version: 1,
+            structural: {
+              contains: [
+                {
+                  type: Cocina::Models::ObjectType.file,
+                  externalIdentifier: "#{druid}_1",
+                  label: 'file1.mp4',
+                  filename: 'file1.mp4',
+                  size: 123,
+                  version: 1,
+                  hasMimeType: 'video/mp4'
+                }
+              ]
+            }
+          }
+        ]
+      }
+    end
+
+    # setup fake caption vtt and json files in the workspace directory which matches the names of the audio and video files in the Cocina
+    before do
+      create_speech_to_text_file('file1_m4a.vtt')
+      create_speech_to_text_file('file1_m4a.json')
+      create_speech_to_text_file('file1_mp4.vtt')
+      create_speech_to_text_file('file1_mp4.json')
+    end
+
+    # rubocop:disable RSpec/ExampleLength
+    it 'runs the update cocina robot and correctly adds the renamed output files to the correct resource' do
+      new_cocina = test_perform(robot, druid)
+      new_audio_json_file = new_cocina.structural.contains[0].structural.contains[1]
+      new_audio_vtt_file = new_cocina.structural.contains[0].structural.contains[2]
+      new_video_json_file = new_cocina.structural.contains[1].structural.contains[1]
+      new_video_vtt_file = new_cocina.structural.contains[1].structural.contains[2]
+      expect(new_cocina.structural.contains.size).to eq 2
+      expect(new_cocina.structural.contains[0].structural.contains[0].filename).to eq 'file1.m4a'
+      expect(new_audio_json_file.filename).to eq 'file1_m4a.json'
+      expect(new_audio_json_file.use).to be_nil
+      expect(new_audio_vtt_file.filename).to eq 'file1_m4a.vtt'
+      expect(new_audio_vtt_file.use).to eq 'caption'
+      expect(new_cocina.structural.contains[1].structural.contains[0].filename).to eq 'file1.mp4'
+      expect(new_video_json_file.filename).to eq 'file1_mp4.json'
+      expect(new_video_json_file.use).to be_nil
+      expect(new_video_vtt_file.filename).to eq 'file1_mp4.vtt'
+      expect(new_video_vtt_file.use).to eq 'caption'
+    end
+    # rubocop:enable RSpec/ExampleLength
+  end
+
   context 'with a txt file' do
     # setup a fake caption txt and json file in the workspace directory which matches the name of the audio file in the Cocina
     before do